New Directions in Breastfeeding Research St. John’s, NL: November 24, 2010

1. Breastfeeding and the prevention of obesity and diabetes: What does the research say about the link? New Directions in Breastfeeding Research St. John’s, NL: November 24, 2010 Patricia J. Martens IBCLC, PhD Director, MCHP; Professor, University of Manitoba CIHR/PHAC Applied Public Health Chair I have no potential conflicts of interest to disclose

2. Objectives • Explain how systematic reviews and meta-analyses look at the existing literature to summarize associations • Explain at least 2 cautions • Summarize findings of relationship between breastfeeding and a reduced risk of overweight and obesity, and comment on possible mechanisms • Summarize findings of relationship between breastfeeding and type 2 diabetes, and comment on possible mechanisms • So what? Relating it to programs/policies at the family, community, provincial and national level

3. Quick Lesson #1: what does p<.05 mean? • p<.05 • The probability of seeing a difference this big, just by chance alone, is less than 5% • so … Statisticians consider this “rare”, so they will conclude that there’s a real difference (i.e., they have found a “statistically significant result”)

4. Quick Lesson #2: Systematic reviews and meta-analyses Systematic review looks at existing literature on a topic, and summarizes it. Authors have strict criteria for inclusion, and check for potential biases. Subset: Meta-analysis does a mathematical calculation to determine the size of an effect if all the studies included were in one BIG study.

5. Why do you need to “combine” studies? • the more people in your study, the less likely that you will make the mistake of concluding there is no effect, when in reality there was • So a meta-analysis combines studies, as if they were one big study! • Less risk of a type 2 error

6. Quick Lesson #3: what is type 1 and type 2 error?

7. Type I error • The “over-enthusiast” error • You conclude there IS a statistically significant finding (alternate hypothesis, H1) even though at the population level you would have found no difference (null hypothesis, Ho) • The p-value tells you about Type I error • p<.05: you’ll make a type I error 5% of the time • Yikes – that’s 1 time out of 20 (important when reading research studies)

8. Type 2 error • The “pessimist” or skeptic error • You conclude “not statistically significant” and stick with the null hypothesis (Ho) even though at the population level you would have found a difference (H1) • Often the result of not enough people in your study (“low power”)

9. Quick Lesson #4: What do RR and OR mean? • RR is a “rate ratio” or “relative risk” • OR is an “odds ratio”

10. A note about Relative Risk (RR) and Odds Ratio (OR) • RR = the relative “risk” of a certain outcome for the exposed group versus the non-exposed group • RR = 3 is three times the risk of getting the disease if you are exposed to the risk compared to not being exposed • RR = 0.8 is only .8 times the risk (ie, less risk) of getting the disease if you are exposed compared to not being exposed • OR = the relative “odds” (odds ratio) • Similar to RR ONLY if the prevalence is small (in which case, you can talk about “three times the risk” etc.) • Odds are NOT the same as risk, but often are talked about in a similar way (but be careful unless it is a rare event)

12. The meaning of RR and OR • For RR and OR, “1” is the null hypothesis – greater than 1 means the risk is increased, less than 1 means it is decreased. • 95% Confidence Intervals • Where you expect to find the true population estimate, 95% of the time. As sample size increases, 95% CI gets smaller (more certain) • RR or OR • If the 95% CI cross over 1, then it’s not statistically significant!! (e.g. 0.8 to 1.2)

13. Mercer AM, Teasley SL, Hopkinson J, McPherson DM, Simon SD, Hall RT. Evaluation of a Breastfeeding Assessment Score in a Diverse Population. J Hum Lact 2010 26: 42-48. Note: how is the p-value and the OR telling you the same thing?

14. Quick Lesson #5: how is a meta-analysis diagrammed? • OR or RR are diagrammed for each selected study in the meta-analysis, showing the 95% CI of each study • If that crosses over 1, then the study concluded “not statistically significant”, BUT … • The combined OR or RR is shown at the bottom, as if it were one big study (and so the 95% CI is much smaller)

15. Arenz et al. 2004

16. So, here we go! • Breastfeeding and protection against obesity or overweight • Four major meta-analyses • Arenz et al. (2004, 2009) – 9 studies • Harder et al. (2005) – 17 studies • Owen et al. (2005) – 28 studies, various studies analyzed for particular subsets • Horta et al. (2007) for the WHO – 33 studies • One major critique of first three (AHRQ Report by Ip et al. 2007)

17. Breastfeeding and protection against obesity or overweight • Who is the AHRQ? • Agency for Healthcare Research and Quality (USA Department of Health and Human Services) • How did they rate the meta-analyses? • Arenz et al. 2004: Grade A • Harder et al. 2005, and Owen et al. 2005: Grade B (suboptimal for potential confounding)

18. Arenz et al. 2004: Grade “A” by AHRQ • Included prospective, cross-sectional and case-control studies; English, Italian, French, Spanish, German; 1966-2003 • had to control for at least 3 “confounders” (birth weight, parental overweight, parental smoking, dietary factors, physical activity, socioeconomic status or parental education) • OR or RR, and age at last followup of 5-18 yr • Feeding mode reported, obesity by BMI percentiles • Found 954 studies, 28 met criteria, 19 not eligible, 9 studies (n=69,000) kept in final calculation

19. Arenz et al. 2004 • No evidence of publication bias or heterogeneity • Comparing any breastfeeding to no breastfeeding, a protective effect on obesity • Crude OR 0.67 (95% CI 0.62-0.73) • Adjusted OR 0.78 (0.71-0.85) • Protective effect of breastfeeding was more pronounced in studies with adjustment for less than 7 potential confounders, but BOTH significant • 0.69 (0.59-0.81) with less confounder adjustment • 0.78 (0.70-0.87) with more adjustment • Arenz 2009: aOR 0.77 (0.72-0.82), similar to previous finding but included more studies in the meta-analysis

20. Arenz et al. 2004 NOTE: comparing never breastfed to ever breastfed

21. Arenz et al. 2004 (page 1253) “It has been reported that breast-feeding does not shift the whole distribution of BMI to the left, but only the upper tail as clearly shown in the publication of Koletzko et al. We are not aware of other studies on the interdependencies of breast-feeding and childhood obesity taking into account both an effect on the median and the upper tail of the distribution. Therefore, we can only hypothesize that the impact on the upper tail of the distribution is the genuine effect of breastfeeding. Interestingly, the obesity epidemic in children as opposed to adults is predominantly caused by an increase of the BMI distribution in the upper percentiles. Therefore, it appears reasonable to consider exposures which affect the upper tail mainly.”

22. Why measure % obese or overweight? (Dewey 2003) • There may be little change in the overall MEAN of the BMI of a population • But there may be dramatic changes in the % in a certain category (like obese)

23. Dewey 2003 (page 11) • “The outcomes must have included the percentage of children who were overweight, not just the mean weight for length or body mass index (BMI: weight (kg)/height (m)2). This criterion is included because our interest is in the right-hand tail of the distribution, not the central tendency [i.e., mean]. It is possible that breastfeeding reduces the extremes at both ends—both overweight and underweight— which would result in a reduced prevalence of overweight but no difference in mean BMI.”

24. Quick Lesson #6: normal distributions and “shifts”

25. The importance of a population perspective on public health Rose's Theorem: "a large number of people at small risk may give rise to more cases of disease than a small number who are at high risk" • Rose, G. The Strategy of Preventive Medicine. Oxford, England: Oxford University Press; 1992.

26. The importance of a population-based approach slide curve over 1/2 a Standard Deviation 31% “healthy” 50% “healthy” LESS healthy MORE healthy

28. Rose-Theorem Coloured Glasses: Population-based Effects! – shift to increase health for all

29. Rose-Theorem Coloured Glasses: “squash” sometimes happens, but we want “squish” to reduce two extreme outcomes of overweight and underweight (especially overweight)

30. Rose-Theorem Coloured Glasses: Population-based Effects! – shift and “squish” to increase health and reduce inequity

31. Harder et al. 2005 (remember – Grade B) • 17 studies, n=120,831 • Must compare breastfed with exclusively formula fed; must report duration of breastfeeding; risk of overweight • Dose-response found • OR 0.94 (0.89-0.98) per MONTH of breastfeeding, lasting up to 9 months duration of breastfeeding (i.e., odds of overweight reduced by 6% per month) • For 9+ months of breastfeeding, the OR is 0.68 (0.50-0.91)

33. Harder et al. 2005 NOTE: comparison group is exclusively formula-fed babies

34. Owen et al. 2005 (Grade B) • 28 studies (n=298,900) • OR = 0.87 (0.85-0.89) • Where info available on breastfeeding duration: • OR 0.81 (0.77-0.84) for breastfeeding 2+ months versus never breastfed

35. Owen et al. 2005

36. Ip et al. 2007 (AHRQ report) • Findings from the three studies suggest that breastfeeding is associated with a reduced risk of obesity in later life.

37. Horta et al. 2007 (WHO) • 33 studies, 39 estimates (some evidence of publication bias); followup from 1 to 66 years • OR=0.78 (0.72-0.84) comparing breastfed to not breastfed for risk of overweight and obesity • Conclude: small protective effect on prevalence of obesity, in spite of evidence of publication bias • (rebuttal from Cope and Allison 2008)

38. Horta et al. 2007 (WHO)

39. Twells and Newhook 2010 • N=1,025 children born in 2001, attending the Newfoundland/Labrador Pre-Kindergarten Health Fair, 10 sites around St. John’s (62% of children attend this). Mean age of child was 4.5 years. 73% were breastfed, 42.6% exclusively breastfed at 3 months • Compares exclusively breastfed for 3 months to non-breastfed babies, controlling for child’s gender, preterm or full term, child age, mother’s education, maternal smoking. • AOR = 0.66 (0.45-0.97).

40. Twells and Newhook 2010

41. So what about contradictory findings? • Kramer et al. 2007 and Kramer et al. 2009 (PROBIT study in Belarus) • No statistically significant difference in BMI between RCT’s intervention and control site births (15.6 kg/m2 in both at age 6 ½ years)

42. Possible reasons for different findings in Kramer et al. • Didn’t really do a study on obesity directly, but rather on the effects of being at a BFHI hospital • Only women who began breastfeeding were enrolled in this cohort (so no formula-fed only for comparison) • Small differences in percentages breastfed between intervention and control sites • Only looked at mean BMI, not percentage obese or overweight

43. Kramer et al. 2009

44. Possible reasons for different findings in Kramer et al. 2007 • So kind of like comparing apples and apples! • Far less problem with overweight/obesity in Belarus compared to Canada • obesity and overweight is higher in the HIGH SES group in Belarus, totally contradictory to what we find in Canada

45. HOWEVER … good observation by Kramer (2009:420S) “But I would like to make a point about the obesity epidemic. If breast-feeding had a potent effect on preventing obesity, would we be living the obesity epidemic at the same time as we have witnessed a renaissance in breast-feeding? When parents started putting their babies on their backs to sleep, the SIDS rate went down; when people stopped smoking, lung cancer rates went down; when people started using seatbelts, deaths from automobile accidents went down. No one ever said that breastfeeding is the ‘‘cure-all’’ preventive measure for obesity. But if it were having a potent effect, would we have seen the epidemic we are witnessing now? Another way of rephrasing that is, whether or not breast-feeding has a small effect in any country, countries such as ours that are experiencing this epidemic have to come up with better ways of controlling it.” • What to say to this? Is breastfeeding maybe mitigating what could be a worse epidemic? Is it so complex that no single intervention shows huge effects?

46. Metzger and McDade 2010 • 2,907 children, but within this, 488 sibling pairs (one breastfed, one not) aged 9-19, recall data of ever breastfed • Controlled for reasons for not breastfeeding, age of mother, SES, birth order, return to work. • Only a small difference (but p<.05) in average BMI, but … • When predicting 85th percentile, breastfeeding was associated with an aOR of 0.59 (p<.01) for whole study, and stronger for sibling study

47. Metzger and McDade 2010

48. Possible mechanisms: Dewey 2003 • Formula fed infants versus breastfed: • higher plasma insulin levels, and prolonged insulin response at 6 • Higher insulin stimulates greater adipose tissue • consume 66% to 70% more protein at 3-6 months, and 5-6 times more at 12 months. • Protein stimulates higher insulin secretion • Lower leptin … leptin is a key regulator of appetite, satiety and body fat • Greater body fat during infancy programs the leptin-dependent feedback loop to be less sensitive to leptin later in life.

49. Ip et al. 2007:62 (AHRQ report) • Mechanisms? • Differences in food composition (breastmilk versus formula): diet-related differences in leptin, ghrelin, insulin-like growth factor etc. differ • Food delivery (breast versus bottle) • Food “lifestyle” (breastfeeding versus formula feeding) • Food behaviour (self-regulation and feeding on demand versus set schedules of feeding of predetermined amounts)

50. Ip et al. 2007:62 (AHRQ report) • Mechanisms? • Breastmilk results in different growth kinetics, with formula-fed infants having higher weight gains • Systematic review of 19 studies in developed countries: formula fed infants weighed 600-650 grams more than breastfed infants at one year old